Multi-Agent Systems

Multi-agent systems are computational structures made up of multiple intelligent agents who can process information, interact with their environment, and with other agents. Multi-agent systems can represent many different real-world systems: transportation, healthcare, and networking are just a few examples.

The agents in MAS must typically meet a number of criteria:

• →

  Self-aware: agents are aware of their own internal state (properties)

• →

  Autonomous: agents determine their own actions, instead of being fully controlled by a central source

• →

  Limited information: agents are aware of their “local” environment (as defined by the model) through their own eyes only, instead of being granted full or perfect knowledge of a system

Agents can have various degrees and structures of “intelligence”, in the form of hard-coded logic, utility-maximizing functions, or neural networks.

Agent-Based Models (ABM) are closely related to Multi-Agent Systems, but are differentiated by their goals. The development of an ABM is typically for explaining or demonstrating certain existing phenomena as a result of collective agent behavior, as opposed to a MAS where the primary goal is designing agents which can solve a specific problem. Because of this, ABM agents are typically simpler.